My AI policy

When you're reading or even talking with anyone online anymore, it's getting to be hard to tell when you're actually interacting with machine-generated text. So I thought I'd address my own rules of engagement, once and for all.

This is written in the form of a warrant canary. If I end up doing any of these things, I'll delete them here.

I have received specialized training in machine learning, including generative AI engineering methodologies. I worked on AI ethics at a corporate level for a Fortune 100 company, and I've been a CTO for an AI startup. As a result, I've done a lot of thinking about what I consider to be ethical use of AI technologies, and what I will and won't do with it.

I have used most major general-purpose large language models (LLMs), including ChatGPT, Claude, Gemini, Llama, Gemma, Qwen, etc. I do not subscribe to any at this time, but occasionally poke around at the free tier to keep pace.

There are various technologies that can be described as "artificial intelligence." I draw a line between specialized AI models and general-purpose models. Specialized models, including many designed for translation and speech recognition, have been in use for decades, and are usually far less expensive to train, as well as more accurate and less susceptible to hallucination, than a general-purpose model attempting the same task. I do my best to avoid using the latter when the former will do.

I have used:

specialized models for transcription and captioning
a general-purpose LLM to create brief summaries of text I wrote myself
a general-purpose LLM to create a public image (as a gag for a post criticizing an AI booster)

I do not use generative AI:

to assist in writing, editing, proofreading or publishing
to brainstorm ideas for writing or designing
to compose or send email, comments or private messages
for notetaking, either live or online
to write or evaluate code, including this site
to meaningfully alter photographic or video content (e.g., to add things that weren't there, or remove things that were)
as a source for factual information, either directly or via AI summaries in search results
to chat